Redesigning the Search Experience 98% of Users Had Abandoned

Overview

Search was the second most-visited feature in CX Cloud, yet 98% of users abandoned it after the first query, unsure whether results even covered what they needed. Redesigning search to show clear entity coverage and a scannable results structure lifted task completion from 2% to 32% after launch.

Contribution

Research, Interaction Design, Prototyping

Duration

6 months

2%→32%

task completion

Daily active users growth

11s→6s

time to find result

Context

CX Cloud's search wasn't helping CX Cloud users find what they needed.

I expected search to find my devices, but it's been slow and often returns irrelevant results. It's faster to find my devices by navigating there directly.

CX Cloud user

The redesigned CX Cloud search: tips in the empty state, recent history, and tabbed results

The result: a search experience users could trust.

Tips guided the first query, history removed the need to retype it, and tabbed results made scanning fast.

How might we

Help CX Cloud users reliably find what they need, right when they need it?

Key Decisions

Turned the empty search box into a guide

Ambiguous empty search box

CX Cloud users arrive at search with a specific object in mind: a device IP, a case number, an asset name. But the search box just said “Search,” with no indication of what it actually covered, so users had no way to know whether their query would work before they typed. The question became how to inform users what they could search for, without forcing them to guess a category first.

First exploration

Scoped filterforces a category choice before users know what they need

5 of 7 CX Cloud users missed the filter completely. They didn't know they had to pick a category before they could search, and only noticed it once it was pointed out.

Shipped

Search tips stateshows users what they can search with search tips

Most CX Cloud users preferred this option: they could see what was searchable before they tried again, instead of guessing.

Search history based on user's activities

As a network engineer, I check on my devices over a range of days to resolve an issue or check the logs.

CX Cloud userNetwork Engineer

CX Cloud users revisit the same pages or devices over a range of days. Their work is managing accounts over time, not looking something up once, so the empty search box made them retype yesterday's search or navigate through the pages again. The opportunity was to learn from that activity and surface what they'd likely need again before they had to ask.

Before

A blank search box that gave CX Cloud users nothing to start with

Search for assets, cases, licenses, coverage, and more...

Shipped

Recent search & viewedsurfaces recent searches and viewed items the moment users click in

Search for assets, cases, licenses, coverage, and more...

Anchoring the empty state in recent activity put the most likely next search one click away. Most CX Cloud users used the pre-populated history to navigate without typing anything. Auto-suggest, matching results as CX Cloud users typed, was the other option on the table, but it needed more engineering time than history did. We shipped history first and scoped auto-suggest into a later phase.

Improved visual hierarchy of search result

Confusing visual hierarchy

The existing results page buried its own structure, one flat list, no visual hierarchy, no room to breathe.

Search results in dropdown

Limited space and accessibility challenge

Search categories

Easy to dismiss

Two link texts with same color

Lack of visual hierarchy

123

Search results that mixed devices, cases, and assets into one stream

Mixed results was the most familiar pattern and tested first. 6 of 7 found it unscannable: category counts too high, sorting arbitrary. It looked complete but wasn't actionable.

Mixed results page showing 16 results for 65 with no scannable structure

Shipped

Tabbed resultsseparates categories into one tab each

Time to find a result: 11s → 6s.

CX Cloud users told us they wanted mixed results, every category in one list. It was the obvious choice on paper: it was the pattern they already knew, so there was nothing new to learn. We tested it first anyway. 6 of 7 couldn't scan it: too much was on screen at once, the category counts were too high to mean anything, and there was no shared way to line up a device against a case against a contract, so the order felt random. Familiar turned out not to mean usable. Tabs scoped each view to one type of result, so sorting finally meant something inside a single category. Time to find a result dropped from 11 seconds to 6. The cost: a CX Cloud user searching a device and its related cases now had to switch tabs once, which they accepted because each view was fast to read.

Impact

Global Search regained users' trust to become their primary way to navigate.

Two releases. The first shipped tips and history. Task completion rose from 2% to 32% in the first month. The second added tabbed results. Daily active users grew 5x in four weeks. Time to result dropped from 11 seconds to 6 on repeat queries. CX Cloud users who had abandoned search were returning to it as their primary navigation path. The tips and history system was adopted by the Cases and Assets teams within six months, product areas that had been building their own empty-state patterns independently.

task completion2% to 32%

Measured in the first month post-launch across moderated and unmoderated sessions.

ease score6.7/7

Single Ease Question.

active user growth5x DAU

Daily active users grew from 10 to 50 within four weeks of the tabbed results release.

Key Decisions